This page presents a small fraction of our laboratory databases that we are allowed to publish for non-commercial, research purposes after receiving a correctly executed license agreement.

These are available for download under a link that appears at the respective section of this webpage. The agreements should be signed by a person who is entitled to take legal obligations on behalf of the institution, to which the requesting person is affiliated (it is typically a Dean or Rector at universities or their legal representatives). Please ensure that the agreements are filled and signed correctly, i.e., in clear, legible print, with no missing or mistaken fields. This allows to simplify the process of sharing the data and lets us avoid exchanging unnecessary e-mails and corrections to the agreements.

Please note that the agreement should be SIGNED by an authorized institutional representative, not by the contact person requesting the data. Requests signed by students or regular staff will not be processed.

The executed agreements should be scanned using a good quality scanner and a soft copy should be sent from a university-affiliated e-mail address to Mateusz Trokielewicz (Mateusz.Trokielewicz@pw.edu.pl).

1. Joint Biometric Dataset -- LivDet 2013 Liveness Detection-Iris -- Warsaw Subset

Unfortunately, this dataset is no longer available due to EU's GDPR regulation.

This dataset gathers 852 images of authentic eyes and 815 images of the paper printouts prepared for those eyes and used to successfully forge an example commercial iris recognition system (i.e., samples used in real and successful presentation attacks).

The dataset was used in the first LivDet-Iris spoofing competition (see livdet.org for more details).

You may be interested in reading a paper describing a training subset of this collection.

2. Joint Biometric Dataset -- LivDet 2015 Liveness Detection-Iris -- Warsaw Subset

Unfortunately, this dataset is no longer available due to EU's GDPR regulation.

This dataset is an extension of the 2013 set and is used in the second LivDet-Iris-2015 spoofing competition (see livdet.org for more details). It gathers 2854 images of authentic eyes and 4705 images of the paper printouts prepared for almost 400 distinct eyes. The photographed paper printouts were used to successfully forge an example commercial iris recognition system (i.e., samples used in real and successful presentation attacks).

You may be interested in reading a paper describing this collection and LivDet-Iris 2015 results.

3. Joint Biometric Dataset -- LivDet 2017 Liveness Detection-Iris -- Warsaw Subset

Unfortunately, this dataset is no longer available due to EU's GDPR regulation.

This dataset is an extension of the 2015 set and was used in the third LivDet-Iris-2017 spoofing competition (see livdet.org for more details). It gathers 5,168 images of authentic eyes and 6,845 images of the paper printouts prepared for approximately 450 distinct eyes. The photographed paper printouts were used to successfully forge an example commercial iris recognition system (i.e., samples used in real and successful presentation attacks). The testing subset includes samples a sequestered "unknown" subset of samples acquired by a different sensor than the one used to collect training set. This is to facilitate a research in open-set presentation attack detection.

The competition results will be presented at IJCB 2017.

4. Warsaw-BioBase-Disease-Iris v1.0

[Dataset Release Agreement]

This collection comprises 603 NIR eye images and 222 corresponding color eye images. Most of the images were acquired for eyes affected by different diseases. NIR images were captured by IrisGuard AD100, and color images were captured by general-purpose Canon EOS 1000D as well as professional Topcon DC3 cameras.

Each class of images (i.e., set of images of any unique iris) is accompanied by an ophthalmological commentary (as metadata). Diseases recognized in this collection include: cataract, acute glaucoma, posterior and anterior synechiae, retinal detachment, rubeosis iridis, corneal vascularization, corneal ulcers, haze or opacities, corneal grafting, iris damage and atrophy, and others.

Paper describing the database and experiments (PDF) is available for download.

5. Warsaw-BioBase-Disease-Iris v2.1

[Dataset Release Agreement]

This dataset is an extended version of the BioBase-Disease-Iris v1.0 collection, consisting of 2996 iris images collected from 115 ophthalmology patients (230 unique irides). These include 1793 NIR-illuminated images and 1203 color images. NIR images were captured by IrisGuard AD100, and color images were captured by general-purpose Canon EOS 1000D as well as professional Topcon DC3 cameras.

Each unique class of images is accompanied by a proper ophthalmological commentary metadata.

We also encourage you to download a paper describing the database and experiments (PDF)

6. Warsaw-BioBase-Smartphone-Iris v1.0

[Dataset Release Agreement]

This is a dataset of visible light iris images collected using the rear camera of an iPhone 5s device. Embedded flash was used for all acquisitions to ensure good quality iris images. The dataset contains 3192 images of 139 distinct irides. To our knowledge this is the only publicly available dataset of iris images of such good quality.

We also encourage you to download those three papers describing the database and experiments:
(Paper 1 PDF) (Paper 2 PDF) (Paper 3 PDF)

7. Warsaw-BioBase-Post-Mortem-Iris v1.0

[Dataset Release Agreement]

This is a unique dataset of iris images collected after a person's death. The images were taken from 5-7 hours post-mortem up to 17 days post-mortem. The dataset comprises 480 NIR-illuminated images obtained using IriShield M2120U iris recognition camera accompanied by 850 color images obtained using Olympus TG-3 consumer camera. The images come from 17 subjects (hence 34 distinct irises). Each case is accompanied by metadata describing age, gender and cause of death

We also encourage you to download those two papers describing the database and experiments:
(Paper 1 PDF) (Paper 2 PDF)

7.1. Warsaw-BioBase-Post-Mortem-Iris v1.1

This is an extended version of the Warsaw-BioBase-Post-Mortem-Iris v1.0 dataset, with the collection protocol extended to almost 34 days post-mortem. The dataset comprises 574 NIR-illuminated images obtained using IriShield M2120U iris recognition camera together with 1023 color images obtained using Olympus TG-3 consumer camera.

Please use the same agreement as the one for v1.0 of the dataset to request access to the v1.1.

8. Warsaw-BioBase-Post-Mortem-Iris v2.0

[Dataset Release Agreement]

This collection of data is an extension of the Warsaw-BioBase-Post-Mortem-Iris v2.0 dataset, and gathers a total of 1200 NIR images and 1797 visible light images collected from 37 post-mortem subjects.

For details see the arXiv preprint of the paper introducing this dataset (accepted for publication in IEEE Transactions on Information Forensics and Security).

8. Warsaw-BioBase-Post-Mortem-Iris v3.0

[Dataset Release Agreement]

This collection of data is an extension of the Warsaw-BioBase-Post-Mortem-Iris v1.1 dataset, and gathers a total of 1094 NIR images and 785 visible light images collected from 42 post-mortem subjects (this set is subject-disjoint with the v2.0 corpus).

Note: the Warsaw-BioBase-Post-Mortem-Iris v3.0 dataset is part of the larger LivDet-Iris 2020 test set. If you plan to test you algorithms with Warsaw's part of the LivDet-Iris 2020 benchmark, use this dataset appropriately to make fair comparisons with the LivDet competition winner (i.e., do not use Warsaw-BioBase-Post-Mortem-Iris v3.0 in training).

For details see the paper introducing this dataset (open-access)

9. Warsaw-BioBase-Pupil-Dynamics v1.0

This dataset is managed by Research & Academic Computer Network NASK.

The dataset is composed of iris segmentation results when the eye is stimulated by visible light. No iris images are provided in this dataset. The data was collected for 54 different eyes. For each eye a pupil size is recorded during 30 seconds of observation under varying light conditions (10 seconds of darkness,5 seconds after step-up change in light intensity and 10 seconds after step-down change in light intensity). The segmentation results are provided by three different algorithms (Czajka, VeriEye and MIRLIN). 25 frames per second are taken.

This dataset has been prepared jointly by WUT and Research & Academic Computer Network NASK – Research Institute in Poland. NASK owns copyright to this set and serves as the only source for this data. NASK holds a US patent on using pupil dynamics for biometric presentation attack detection (US Patent No. 8,061,842).

You may be interested in reading papers describing the database and baseline results (IEEEXplore, Springer).

10. Warsaw-BioBase-Pupil-Dynamics v2.1

Warsaw-BioBase-Pupil-Dynamics v2.1 dataset is managed by Research & Academic Computer Network NASK.

This dataset is an extension of the Warsaw-BioBase-Pupil-Dynamics v1.0 set and is composed of iris segmentation results (calculated by the OSIRIS software) when the eye is stimulated by visible light. The data was collected for 166 different eyes of 86 subjects. For each eye a pupil size is recorded during 30 seconds of observation under varying light conditions (10 seconds of darkness,5 seconds after step-up change in light intensity and 10 seconds after step-down change in light intensity). 25 frames per second are taken.

This dataset has been prepared jointly by WUT and Research & Academic Computer Network NASK – Research Institute in Poland. NASK and WUT own copyright to this set and serve as the only source for this data. NASK holds a US patent on using pupil dynamics for biometric presentation attack detection (US Patent No. 8,061,842).

This data has been used in evaluations presented in the Chapter "Application of Dynamic Features of the Pupil for Iris Presentation Attack Detection" to appear in "Handbook of Biometric Anti-Spoofing (2nd Edition)" and will be available when the book is published.

10. Warsaw-BioBase-Pupil-Dynamics v3.0

[Dataset Release Agreement]

This dataset is composed of iris videos collected from 42 individuals following the protocol given in the paper linked below: for each eye a video is recorded during 30 seconds of observation under varying light conditions (10 seconds of darkness, 5 seconds after step-up change in light intensity and 10 seconds after step-down change in light intensity). 25 frames per second are taken.

For details see the arXiv preprint of the paper introducing this dataset (accepted for the ICB2019 conference).

This dataset has been prepared by WUT, which owns copyright to this set and serve as the only source for this data. Research & Academic Computer Network NASK – State Research Institute in Poland holds a US patent on using pupil dynamics for biometric presentation attack detection (US Patent No. 8,061,842).

11. Warsaw-BioBase-HorseIris v1.0

[Dataset Release Agreement]

This is a novel dataset of iris images collected from horses using a near-infrared 60fps camera from 28 animals, including Arabian race horses. There are 55 classes (irises) available, and each class is represented by approximately 2000 images. There are at least two acquisition sessions available for each class of images.

We also encourage you to download those two papers describing the database and experiments (Paper 1 PDF) (Paper 2 PDF)

12. BioBase-Hand-Thermal v1.0

[Dataset Release Agreement]

This is a dataset of hand thermal images collected in three different sessions using thermal camera FLIR. The dataset contains 21,000 images of 70 subjects (140 classes). To our knowledge this is the first known database of hand thermal maps acquired by thermal sensor in unconstrained scenario.

The description of the database and baseline results are included in our paper: ,,Unconstrained Biometric Recognition based on Thermal Hand Images" (PDF).

Supplementary materials for papers

1. Data-Driven Segmentation of Post-Mortem Iris Images

IWBF 2018 paper

Source codes, network weigths, and manual segmentation results for the experiments described in the paper (PDF) can be obtained from here [link to zip archive].

2. Segmentation of thermal spectrum hand images with a pre-trained off-the-shelf DCNN model

BTAS 2019 paper

Source codes, network weigths, and manual segmentation results for the experiments described in the paper (PDF) can be obtained from here [link to zip archive to appear upon paper acceptance].

2. Post-Mortem Iris Recognition Resistant to Biological Eye Decay Processes

WACV 2020 paper

Source codes, network weigths, trained iris-specific filters for the experiments described in the paper (ArXiv preprint) can be obtained from here can be obtained from here [link to zip archive].